Large-Scale Language Models
摘要
A large-scale language model generally refers to models based on Transformer architecture, such as BERT, GPT-X (X \(=\) 1, 2, 3, 3.5, 4), BART, T5, and Llama. Large-scale language models are fundamentally based on pre-training through self-supervised learning and fine-tuning for specific tasks. The aim of pre-training is to acquire general knowledge about language, and compared to pre-training, fine-tuning requires less data.